Integrating Robotic Sensor and Effector Capabilities with Multi-agent
Organizations
Eric Matson, Scott DeLoach
Multi-Agent and Cooperative Robotics Lab
Department of Computing and Information Sciences
Kansas State University
Manhattan, KS 66506
matson,[email protected]
Abstract
Robots possess many effectors and sensors of various capability. It is often difficult, not only to integrate these numerous capabilities, but also to organize them to ac-complish a set of goals or tasks. Even more difficult is how to reorganize on the condition that a sensor/effector is damaged or incapacitated during operation. In this pa-per, we show how a organization-based multi-agent sys-tem (OMAS) adapts to control the sensor and effectors within a robot. The OMAS can also be extended to model the sensor and effectors for a team of cooperative robots. The use of integrating sensors/effectors via a OMAS will produce a self-reorganizing, fault tolerant control archi-tecture capable of allowing a robot, or a team of robots, to overcome obstacles and faults within dynamic domains where sensor and effector failure rates may be high. An-other positive effect of an OMAS is simplification of sys-tem complexity where a non-trivial number of sensors and effectors exist.
Keywords: Multiagent Systems, Organization
1. Introduction
A common obstacle with robotics is the development of control programs that allow sensors and effectors to work in harmony to accomplish a wide range of tasks. As the demands of the environment become more com-plex and the number of goals increase, the sensors and effectors required to meet the goal and environmental demands will also grow. Growth in sensors and effec-tors will increase the interaction required to be man-aged by the robotic control software. A way of manag-ing and organizmanag-ing the growth of the robotic control is through a multiagent system extended with an
organi-zation theoretic model. In our research, we have devel-oped an organization-based MAS (OMAS) model that can be applied to the problem of sensor and effector organization for a robot or teams of robots. We pro-pose that the use of an OMAS will significantly reduce the complexity of sensor and effector interaction and provide a model to create an efficient self-reorganizing team.
Figure 1. Complexity of Control
Complexity is a common problem in robotics re-search and implementation. The number of sensors and effectors for non-trivial systems can be quite large and, therefore, very complex. As related in Fig. 1, as the number of sensors and effectors increase, the complex-ity of the software control system will increase at an increasing rate. While this curve will be differentiated by the number, type and complexity of each individual type, it merely represents the general trend of com-Proceedings of The 2004 International Conference on Artificial Intelligence (IC-AI'04)
plexity. Reduction of the complexity, or simply man-aging and controlling it, is potentially gained using an organization-based multiagent system. An organization based model, using capability, enables a robot’s sen-sors and effectors to be modularized, which allows ef-fective relationships to be constructed between them. The integration of the organization model additionally allows for specialization of labor around roles that re-duces the relationship complexity shortcoming of many multiagent peer-to-peer systems.
The goal of this research is to show the viability of applying organizational models and MAS for use in robotics. Specifically, we want to show that the result-ing organization-based multiagent systems (OMAS) are a highly useful and functional alternative to tradi-tional teamwork schemes and formalisms [1] [2]. Com-plimentary work in this area has proposed the use of networked robotics without a self-reorganizing multi-agent concept [3]. Our research takes into considera-tion fault tolerant systems and architectures that deal with detecting and handling sensor failure and faults [4], and calibration of sensors to adapt to unknown en-vironmental conditions [5]. Our model tolerates faults by managing the available hardware sensors as a coop-erative, intentional system, focusing on managing their entire set of capabilities instead of simple ”brute force” approach to sensor switching in cases of failure.
This paper defines our organization model in Sec-tion 2. SecSec-tion 3 details the applicaSec-tion of the OMAS model to a robotic instance with a non-trivial num-ber of sensors and effectors. Section 4 provides conclu-sions for this model and Section 5 details future direc-tions and plans.
2. Organization
To implement teams of autonomous, heterogeneous robots, we created an organizational model, which de-fines and constrains the required elements of a stable, adaptable and versatile team. While most people have an intuitive idea of what an organization is, there are no standard definitions. However, in most organizational research, organizations have typically been understood as including agents playing roles within a structure in order to satisfy a given set of goals. Our proposed or-ganizational model (O) contains a structural model, a state model and a transition function. Fig. 2 shows the combined structural and state models using standard UML notation.
O=< Ostructure, Ostate, Otransition>
2.1. Structure
The structural model includes a set of goals (G) that the team is attempting to achieve, a set of roles (R) that must be played to attain those goals, a set of ca-pabilities (C) required to play those roles, and a set of rules or laws (L) that constrain the organization. The model also contains static relations between roles and goals (achieves), roles and capabilities (requires), indi-vidual roles (related), goals and subgoals(subgoals) and a unary relation for conjunctivity between subgoals of a goal(conjunctive). Formally, we model the organiza-tion structure as a tuple:
Ostructure=< G, R, L, C, achieves, related, requires,
subgoal, conjunctive > where: achieves:R, G→[0..1] related:R, R→Boolean requires:R, C→Boolean subgoal:G, G→Boolean conjunctive:G→Boolean
Figure 2. Organization Model
The team goals include the goal definitions, goal-subgoal decomposition, and the relationship between the goals and their subgoals, which are either conjunc-tive or disjuncconjunc-tive. Roles define parts or positions that an agent may play in the team organization. In gen-eral, roles may be played by zero, one, or many agents simultaneously while agents may also play many roles at the same time. Each role requires a set of capabil-ities, which are inherent to particular agents and may include sensor capabilities (sonar, laser, or video, etc.), actuator capabilities (movement type, grippers, etc.),
or computational capabilities (processing power, algo-rithms, communications, etc.). Robots are unique in the area of capabilities versus software agents; robot’s physical capabilities may improve or degrade over time, which can often cause the team to reorganize. Orga-nizational rules are used to constrain the assignment of agents to roles and goals within the organization. Generic rules such as “an agent may only play one role at a time” or “agents may only work on a single goal at a time” are common. However, rules are often ap-plication specific, such as requiring particular agents to play specific roles. The structural model relations define mappings between the structural model compo-nents described above. A role that can be used to sat-isfy a particular goal is said to achieve that goal, while a role requires specific capabilities and may work di-rectly with other roles, thus being related to those roles. Achieves is modeled as a function to capture the rela-tive ability of a particular role to satisfy a given goal.
2.2. State
The second element of the Organization Model is its state. The Organizational State (Ostate) is an instance
of the organizational structure at a point in time. As the Organization Model is a template, the state is an in-stance of the model. In an inin-stance of an organization state, each of the elements will be bound to a set of val-ues that represent the organization attributes. An or-ganization will possess at least one goal, one role to ac-complish the goal, and one agent to play the role where the agent will possess capabilities required by the role. Not every organization state element is required to be populated by an instance variable for creation of a valid organization. The constraints and laws of an organiza-tion will govern the requirements of a specific state.
The organizational state model defines an instance of a team’s organization and includes a set of agents (A) and the actual relationships between the agents and the various structural model components.
Ostate=< A, possesses, capable, assigned, coord >
where:
possesses:A, C→[0..1]
capable:A, R→[0..1]
assigned:A, R, G→[0..1]
coord:A, A→Boolean
An agent that possesses the required capabilities for a particular role is said to be capable of playing that role. Since not all agents are created equally, possesses is modeled as a real valued function, where 0 would rep-resent absolutely no capability to play a role while a
1 indicates an excellent capability. In addition, since agent capabilities may degrade over time, this value may actually change during team operation. The ca-pable function defines the ability of an agent to play a particular role and is computed based on the capabili-ties required to play that role. During the organization process, a specific agent is selected to play a particu-lar role in order to satisfy a specific goal. This relation-ship is captured by the assigned function, which in-cludes a real valued score that captures how well an agent, playing a specific role, can satisfy a given goal. When an agent is actually working directly with an-other agent, it is coordinating (coord) with that agent. Thus, the state model defines the current state of the team organization within the structure provided by the structural model.
2.3. Transition
The Organization Transition Function (Otransition)
defines the ability of the organization to reorganize from an instance state to the next instance state over the organization life span. From the initial organiza-tion, through its terminaorganiza-tion, the organization may transition its state model numerous times.
The organization transition function defines how the organization may transition from one organizational state to another over the lifetime of the organization. Since the team members (agents) as well as their indi-vidual capabilities may change over time, this function cannot be predefined, but must be computed based on the current state, the goals that are still being pursued, and the organizational rules. In our present research with purely autonomous teams, we have only consid-ered reorganization that involves the state of the orga-nization. However, we have defined two distinct types or reorganization: state reorganization, which only al-lows the modification of the organization state, and structure reorganization, which allows modification of the organization structure (and may require state reor-ganization to keep the orreor-ganization consistent). To de-fine state reorganization, we simply need to impose the restriction that:
Otrans(O).Ostructure=O.Ostructure
Technically, this restriction only allows changes to the set of agents, A, the coord relation, and the pos-sesses, capable, and assigned functions. However, not all these components are actually under the control of the organization. For our purposes, we assume that agents may enter or leave organizations or relation-ships, but that these actions are triggers that cause
reorganizations and are not the result of reorganiza-tions. Likewise, possesses (and thus capable as well) is an automatic calculation on the part of an agent that determines the roles that it can play in the or-ganization. This calculation is totally under control of the agent (i.e. the agent may lie) and the organiza-tion can only use this informaorganiza-tion in deciding its orga-nizational structure. Changes in an agent’s capabilities may also trigger reorganization. That leaves the two el-ements that can be modified via state reorganization:
assigned and coord. Thus, we define state reorganiza-tion as:
Otrans(state):O→O
where
Otrans(state)(O).Ostruct=O.Ostruct
Otrans(state)(O).Ostate.A=Ostate.A
Otrans(state)(O).Ostate.possesses=Ostate.possesses
Otrans(state)(O).Ostate.capable=Ostate.capable
Organization and Reorganization Processes The initial step in organizing an multi agent or robotic team is to use the existing information production goals to es-tablish the organizational roles required to produce the appropriate information. At the same time, the team of agents or robots making up the team must assess their individual and collective capabilities to determine whether they can fulfill the required roles [MD03]. If the required roles can be filled, then the capabilities exist to satisfy the information production goals and the team assigns the necessary roles to agents (effec-tively defining the state of the team’s organization). Once the assignments are made, the team may initi-ate action to satisfy the team information production goals.
Ostate(0)→Ostate(1)
The reorganization process follows the same basic steps as the organization process; however, it differs in the point of initiation. Reorganization is initiated by a trigger event, such as sensor loss, during the execu-tion of an already existing organizaexecu-tion. When such an event occurs, the team must determine if it still has the capabilities to satisfy team information production goals or whether it must reorganize to do so.
Ostate(0)→Ostate(n+1)
Organizational Outcomes The outcomes of the organi-zation and reorganiorgani-zation processes are equivalent. The three available outcomes are goal satisfaction, goal re-laxation or goal abandonment. Goal satisfaction
indi-cates that the capabilities exist within the remaining team to accomplish all critical and non-critical goals. Goal relaxation indicates that capabilities exist within the remaining team to meet all critical goals, but some or all non-critical goals may not be met and will have to be “relaxed”. Goal abandonment indicates the re-maining member’s capabilities do not allow the orga-nization to continue because not all critical goals can be satisfied and success is not achievable.
2.3.1. Capability Robots are defined by the phys-ical and computational capabilities they possess. The robot’s capabilities define the roles they can play in meeting a team goal. For robots, there are two levels of capabilities: computational and physical. The com-putational capabilities are defined by the level of intel-ligence built into the robot. The physical capabilities are defined by the range of sensors and effectors in-cluded as part of the robot’s design. In this research, we will focus on modeling the function of the physi-cal sensors with the capabilities of agents.
So far, we have used the term capability generically. However, we must define it more precisely to demon-strate how we utilize it to integrate robot sensors and effectors. A capability’s existence is based on the col-lective sense in which it is viewed. To specify this, we further define capabilities in relation to agent and roles that exist within a self-reorganizing multiagent team. As described above, an agent possesses specific capa-bilities while roles require particular capacapa-bilities, each with specific scores.
The capability set of an agent,Ca, varies from the
empty set, if the agent possesses no capability, to a complete set of the capabilities that the agent intrinsi-cally possesses. Typiintrinsi-cally, although not always, even a simple agent has multiple capabilities.
Ca(a) ={c|possesses(a, c)>0}
Likewise, the capability set of a role,Cr, is the set of
capabilities required to play that specific role. The ca-pability set formally describes what capabilities are re-quired for agents potentially to enact and play the role [7]. All non-trivial roles must have at least one capa-bility in order to accomplish some task or goal.
Cr(r) ={c|requires(r, c)}
An agent is capable of playing a role if Cr(r) ⊆ Ca(a). How well agent acan play roleris determined by therole capability function (rcf ) that is part of each role definition. Thercf is part of the role and defines a role-specific computation based on the capabilities pos-sessed by an agent. If an agent does not possess one of
the required capabilities, then the agent has no capac-ity to play that role and r.rcf(a) = 0. Thus, the ca-pability score of an agent playing a particular role is defined by:
capable(a, r) =r.rcf(a)
During the organization process, a specific agent is se-lected to play a particular role in order to satisfy a spe-cific goal. This relationship is captured by the assigned function, which includes a real valued score that cap-tures how well an agent, playing a specific role, can sat-isfy a given goal. This score is computed by
assigned(a, r, g).score=capable(a, r)∗achieves(g, r) When an agent is actually working directly with an-other agent, it is coordinating(coord)with that agent. Thus, the state model defines the current state of the team organization within the structure provided by the structural model.
The agents that form a team have a collective ca-pability. Similarly, the set of roles required to achieve the overall organizational goal also have a set of re-quired capabilities. We define these as team capabili-ties,CA, and required capabilities,CR.
CA(O) ={c| ∃a:A•c∈Ca(a)}
CR(O) ={c| ∃r:R•c∈Cr(r)}
To form a viable organization, these sets must be min-imally overlapping such that the capabilities required are contained in the capabilities available from the agents, with respect to the organization, such that:
CR(O)⊆CA(O)
3. Robotic Implementation
In this section, we will use a robotic instance to demonstrate how the organization-based concept and how formalizations can be used to integrate sensors and effectors.
An important measure of a robot is its physical abil-ities, each with a specific set of capabilities to play a role within an organization. Whereas a robot is defined by the combination of its computational and physical characteristics and capabilities, we will focus on the physical attributes; sensors and effectors. Each sensor and effector binds to a capability and the capabilities tie to an agent contained within an organization. Our demonstration robot is a Nomad Scout [8] as shown in Fig. 3. The effectors of the Nomad Scout are triv-ial, but the use of the sensors, as shown in Figure 4,
Figure 3. Nomad Scout
are sufficient to capture and apply to organization the-oretic concept.
An integral concept in the development of an orga-nization, to satisfy a set of goals, is the specialization of labor. In our case, the specialization is captured by the roles defined to accomplish the given goals. Spe-cialization will allow the organization to avoid com-plex individual coordination but it will call for coor-dination between individuals[10]. Specialization, using roles, limits the interactivity to only the agent peers that must communicate by definition. The organiza-tion model formalizes this by the coordinaorganiza-tion between agents coord: A, A →Boolean. The reduction of co-ordination, between all individuals, will allow us to re-duce the increase in complexity as the number of indi-vidual sensors grow. In the case of the Nomad Scout robot, we no longer need to work with 22 different sen-sors, we will now organize around two roles that require the capability to manage a sonar or a tactile bump, re-spectively.
The Nomad Scout robot has sonar and tactile bump sensors. Collectively, there are 22 sensors consisting of 16 sonar sensors and 6 tactile bump sensors. The sonar ring is a Sensus 200 consisting of 16 Polaroid 6500 sonar ranging modules fixed in 22.5◦ increments in a full 360◦ configuration. The Polaroid 6500 mod-ule can accurately measure distances from 6 inches to 35 feet,±1%. There are six bump sensors configured on the front and rear arcs of the robot. The bump sen-sors are tactile, so physical contact must be made with a physical object to be enacted. The bump sensors, un-like the sonar, do not provide a 360◦range of detection,
as shown by the graphical comparison of the sonar and bump sensor configurations.
Example: Environment Scanning In this example, we show how the organization model can be used in a
sim-Figure 4. Nomad Sensors
ple scenario, stationary environmental scanning, with a single Nomad Scout robot. To create the organization for integrating the Nomad’s sensor and effector capa-bilities, goals, roles and agents must be created that capture the essence of the specified task. For integra-tion of sensor capabilities, the structural model must be fully developed. The organization structural element
Ostructureis captured by:
Organization:Scan the environment
Goal={scan22.5◦, bump, integrate, manage,
scan360◦}
Role={scannersonar, scannerbump, integrator,
manager} Law={}
Capability={detect22.5◦, detecttouch,
integratesonar, integratebump, manageintegrator}
T he achieves relation:
achieves(scannersonar, scan22.5◦)→1
achieves(scannerbump, bump)→1
achieves(integrator, integrate)→1
achieves(manager, manage)→1
T he related relation:
related(scannersonar, integrator)→true
related(scannerbump, integrator)→true
related(integrator, manager)→true T he role capabilities required are as f ollows:
Cr(scannersonar) ={detect22.5◦}
Cr(scannerbump) ={detecttouch}
Cr(integrator) ={integratesonar, integratebump}
Cr(manager) ={manageintegrator}
T he subgoal relation:
subgoal(scan22.5◦, scan360◦)→true
subgoal(bump, scan360◦)→true
subgoal(integrate, scan360◦)→true
subgoal(manage, scan360◦)→true
T he conjunctive relation:
conjunctive(scan360◦)→true
To complete the organization theOstatewill need to be
defined for this scenario. Thestateis defined byOstate
where the following relationships form:
T he agent def initions f or this scenario are:
A={agent1, agent2. . . agent25}
T he capabilities possessed by each agent are:
Ca(agent1). . . Ca(agent16) ={detect22.5◦}
Ca(agent17). . . Ca(agent22) ={detecttouch}
Ca(agent23) ={integratesonar}
Ca(agent24) ={integratebump}
Ca(agent25) ={manageintegrator}
T he capabilities f or an agent to play a role are:
capable(agent1, scannersonar). . .
capable(agent16, scannersonar)→1
capable(agent17, scannersonar). . .
capable(agent25, scannersonar)→0
capable(agent1, scannerbump). . .
capable(agent16, scannerbump)→0
capable(agent17, scannerbump). . .
capable(agent22, scannerbump)→1
capable(agent23, scannerbump)→0
capable(agent24, scannerbump)→0
capable(agent25, scannerbump)→0
capable(agent1, integrator). . .
capable(agent22, integrator)→0
capable(agent23, integrator)→1
capable(agent24, integrator)→1
capable(agent25, integrator)→0
capable(agent1, manager). . .
capable(agent24, manager)→0
capable(agent25, manager)→1
T he coordinating agents are:
coord(agent1, agent23). . .
coord(agent16, agent23)→true
coord(agent17, agent24). . .
coord(agent22, agent24)→true
coord(agent23, agent25)→true
coord(agent24, agent25)→true
The assignment of agents, roles and goals completes the set of relations required to form an organization.
In this case, there are singular agents capable of play-ing each role and thus satisfyplay-ing each of the main goals conjunctive subgoals. While there are trivial ment alternatives for this organization, such as assign-ing agent2 to a different sonar unit, overall the
re-sulting organization instanceOstate(1) or alternatively,
Oscan22.5◦ is by default optimal. If there are assign-ment alternatives, the organization computation score
assigned(a, r, g).score will be employed to determine the optimal organization instance based on the capa-bility of an agenta to play a roler, where the role r
satisfies some goalg.
T he assignments are:
assigned(agent1, scannersonar, scan22.5◦). . .
assigned(agent16, scannersonar, scan22.5◦)→1
assigned(agent17, scannerbump, bump). . .
assigned(agent22, scannerbump, bump)→1
assigned(agent23, integrator)→1
assigned(agent24, integrator)→1
assigned(agent25, manager)→1
In this scenario the capability constraint is satisfied as CR(O) ⊆CA(O) is satisfied. All CR are possessed
byCA:
CR(O) ={detect22.5◦, detecttouch, integratesonar,
integratesonar, manageintegrator}
CA(O) ={detect22.5◦, detecttouch, integratesonar,
integratesonar, manageintegrator, coordinate}
Because the constraints required to form a valid or-ganization have been satisfied, the initial oror-ganization will be instantiated with the given assignments based on the goals, roles and agents.
4. Conclusions
In this paper, we show how our organization theo-retic model can be applied to the problem of sensor and effector integration. By using an organization model, intrinsic capabilities of sensors and effectors can be en-abled to cooperatively work together. A second impor-tant feature allows growth in the number of sensors and effectors without exponential growth in the com-plexity to control the sensors.
Advantages There are many advantages in using or-ganization models. Application to many domains, the ability to limit or reduce complexity, reduction of over-all necessary agent interaction, assumed intent, coop-eration, fault tolerance and recovery are just a few of the advantages. Our model is generic, and purposely
designed with the intent of making it applicable in a number of problem domains. With the ability to spe-cialize labor, the complexity can be reduced by the na-ture of simple planning and interaction in relation to roles that will satisfy the system goals. Reducing com-plexity and specializing roles will reduce overall inter-action between agents in comparison with strict peer point-to-point agent implementations. This complex-ity reduction will allow us to construct larger robotic teams, with less development effort, than is currently possible. Because the organization is created around a specific set of goals, the system becomes intentional by nature, and cooperative by design. Our organiza-tional model allows the transition to a new organi-zation if there is a failure by a agent working within the team. Reorganizing also can be triggered by a sub-optimal state occurring or developing in the current organization. Self reorganization allows for the imple-mented systems to be fault tolerant and recoverable.
Limitations Even though the organization model is de-signed generically, with many domains considered, it will have to be applied to the specific domain problem effectively. If the model is not applied thoughtfully and correctly, the expected results may be less than the ef-fective level desired. We are continuing to further refine the model to capture all domains where organization is needed or can be applied. Tools that guide the plan-ner through the process of developing, asking the valid questions and capturing all necessary organization in-formation will reduce these issues significantly.
Possible Applications Because of the generic design, our model is enabled for use in a wide range of ap-plications. As of the time of submission, we have used the model in the Adaptive Information Systems (AIS) [12], robotics, and agent-oriented software engineer-ing domains. As we have shown here, there are many goal-based robotic applications where our organization model can be applied using an OMAS approach.
5. Future Work
This work is part of a larger effort to more fully de-fine the usefulness of an organizational theoretic ap-proach to building a multi-agent system. In the near future, we plan to add new sensor types and thus as-sign more, different types of agent capabilities. This will allow us to more fully evaluate the scalability of the organizational model and the effectiveness of our organizational reasoning techniques. We will also inte-grate the organization model into the Multiagent Soft-ware Engineering (MaSE) tool agentTool[13]. This will assist in the construction of valid and useful
organiza-tion based applicaorganiza-tion while reducing the problems in designing non-trivial organizations from scratch.
6. Acknowledgments
This work is sponsored by the Air Force Office of Sci-entific Research (AFOSR) under grant number F49620-02-1-0427.
References
[1] Tambe, M. and Zhang, W., Towards Flexible Teamwork
in Persistent Teams. Second International Conference on Multi-Agent Systems, 1996.
[2] Far, B.H., Hajji, H., Saniepour, S., Soueina, S.O.,
Elkhouly, M.M., Formalization of Organizational Intel-ligence for Multiagent System Design. IEICE Transac-tions on Information and Systems. Volume E83-D, No. 4, April 2000.
[3] McKee, G., Schenker, P., Baker, D. Networked Robotics
Concepts for Space Robotics Systems, Proceedings of the Robosphere 2002 Workshop, NASA Ames Research Center, Moffett Field, California, November 14-15, 2002.
[4] Roumeliotis, S., Sukhatme, G., Bekey, G., Sensor Fault
Detection and Identification in a Mobile Robot, Pro-ceedings of 1998 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS ’98), Victo-ria, Canada, Oct. 13-17, 1998, pp.1383-1388.
[5] Soika, M. Grid Based Fault Detection and Calibration of
Sensors on Mobile Robots. Proceedings of the 1997 IEEE International Conference of Robotics and Automation (ICRA ’97). Albuquerque, New Mexico. April 1997.
[6] Matson, E., DeLoach, S. Organizational Model for
Co-operative and Sustaining Robotic Ecologies, Proceed-ings of the Robosphere 2002 Workshop, NASA Ames Research Center, Moffett Field, California, November 14-15, 2002.
[7] Dastani, M., Dignum, V., Dignum, F. Role Assignment
in Open Agent Societies. Proceedings of the Second Joint Conference on Autonomous Agents and Multiagent Sys-tems (AAMAS ’03), Melbourne, Australia, July 14-18, 2003.
[8] Nomadic Technologies, Inc. July 12, 1999. Nomad Scout
User’s Manual, Mountain View, CA
[9] Matson, E., DeLoach, S. Using Dynamic Capability
Evaluation to Organize a Team of Cooperative, Au-tonomous Robots, Proceedings of the 2003 International Conference on Artificial Intelligence (IC-AI ’03), Las Ve-gas, Nevada, June 23-26, 2003.
[10] Houthakker, Henrik, S., Economics and Biology: Spe-cialization and Speciation, Kyklos, 1956, 9,(2), 181-187. [11] DeLoach, S., Matson, E., Li, Y. Exploiting Agent Ori-ented Software Engineering in the Design of a Coopera-tive Robotics Search and Rescue System. The Interna-tional Journal of Pattern Recognition and Artificial In-telligence,17 (5), pp. 817-835, August 2003.
[12] Matson, E., DeLoach, S. Organization-based Adaptive Information Systems for Battlefield Situational Anal-ysis, Proceedings of the IEEE International Conference on Integration of Knowledge Intensive Multi-Agent Sys-tems (IEEE KIMAS ’03), Boston, MA, October 1-3, 2003.
[13] DeLoach, S., Wood, M. Developing Multiagent Systems with agentTool. in Intelligent Agents VII. Agent The-ories Architectures and Languages, 7th International Workshop ( ATAL 2000, Boston, MA, USA, July 7-9, 2000), C. Castelfranchi, Y. Lesperance (Eds.). Lecture Notes in Computer Science. Vol. 1986, Springer Verlag, Berlin, 2001.
